Hmm...
Couldn't they mean that there is a 3% chance of the real result being outside a single standard deviation radius from the mean? Or something completely different? They have, indeed, not given us any notion of what it is that they consider an error.
Could they mean that the standard deviation is 3% of the votes?

Problem is, that when you are referring to elections, there are at least two percentages you can be referrring to: those regarding probability, and those regarding the actual percentage of public support each candidate had received.